AI-Enhanced and Automated Indirect Process Monitoring at the Sensor Edge

This work presents a novel concept of the edgebased indirect measurement framework designed for real-time process automation, leveraging inverse problem-solving methodologies and AI-driven inference. We discuss the potential of this approach in sample applications such as automated clothing segregation using near-infrared (NIR) sensors and measurement of sugar concentration in water using capacitive micromachined ultrasonic transducers (CMUTs). The paper highlights the role of stochastic computing implemented on FPGAs to enhance efficiency and enable low-latency processing directly at the sensor edge. We provide an in-depth analysis of why stochastic computing and inverse neural operators are promising for future real-time AI/ML hardware acceleration in indirect measurement applications. Future work will focus on practical implementation and validation of these concepts through FPGA-based inference models.